The Best Paper Award
Pattern Recognition with Gaussian Mixture Models of Marginal Distributions
Masako Omachi　・　Shinichiro Omachi
（英文論文誌D　平成23年2月号掲載）
　In the field of statistical pattern recognition, precise estimation of data distribution is important for achieving high recognition accuracy. When a large amount of samples can be obtained via web, this estimation will be an easy problem because we can use it for precise estimation of data distribution. However, there are many real-life problems that are difficult to collect an enough amount of samples, so precise estimation of data distribution with a small number of samples is a still challenging problem.
　 In this paper, a simple method is proposed for estimating multimodal data distribution based on the Gaussian mixture model. In this method, multiple random vectors are generated after classifying the elements of the feature vector into subsets so that there is no correlation between any pair of subsets. The Gaussian mixture model for each subset is then constructed independently. As a result, the constructed model is represented as the product of the Gaussian mixture models of marginal distributions. To make the classification of the elements effective, a graph cut technique is used for rearranging the elements of the feature vectors to gather elements with a high correlation into the same subset.
　 The proposed method is applied to the character recognition problem called MNIST that requires high-dimensional feature vectors. Experimental results show that the proposed method improves the accuracy of classification with a small amount of computation time for constructing the Gaussian mixture models. In addition, the effect of classifying the elements of the feature vectors is shown by visualizing the distribution. We concluded that this paper has a good reputation because the proposed method can be regarded as a general-purpose technique in the field of statistical pattern recognition.

Close